This section provides background information relating to the present disclosure, which is not necessarily prior art.
In scanning a document with an overhead scanner or shooting a document with a configured camera or smart phone, a user normally presses the document with a finger during the scanning process or the shooting process. Thus, a finger image of the user exists in a scanned/shoot image, resulting in unpleasantness of the scanned/shoot image of the document. Therefore, there is a need of removing the finger image from the image. In general, for removing the finger image, a finger area needs to be detected firstly, and then the finger area is removed.
The issue of automatic detection of a finger can be considered as an issue of image segmentation. In some skin-color segmentation methods, different color expressions (such as RGB, HSV or YCbCr) and classification methods (such as the Bayes classifier and the Gaussian mixture model) are applied to segment a skin-color area. However, a predefined skin-color model cannot express effectively the color variation in different conditions. By contrast, an adaptive skin-color model may better express the color variation in different illumination conditions, leading to a better segmentation. But if a background color is similar to a skin-color, it is difficult to segment a finger area from a background only with color information. In addition, if a pixel with a color similar to the skin-color exists in a background area, the pixel interferes with a representation of a constructed skin-color model.